Deep Learning and Model Independence

TL;DR

This paper discusses the application of deep learning techniques, especially autoencoders and unsupervised learning, to promote model-independent approaches in high energy physics research, aiming to reduce biases in new physics searches.

Contribution

It introduces novel uses of deep learning for model-independent measurements and proposes a nuanced definition of model independence in physics research.

Findings

Deep learning enables minimally-biased measurements in high energy physics.

Autoencoders and unsupervised methods are effective for model-independent analysis.

Model independence exists in degrees, not as an absolute concept.

Abstract

The lack of evidence in favor of any new physics models means that the search for new physics beyond the Standard Model (BSM) is wide open, with no direction clearly more promising than any other. This marks a turn towards what can be called `model-independent' methods-strategies that reduce the influence of modelling assumptions by performing minimally-biased precision measurements, using effective field theories, or using Deep Learning methods (DL). In this paper, I present the novel and promising uses of DL as a primary tool in high energy physics research, highlighting the use of autoencoder networks and unsupervised learning methods. I advocate for the importance and usefulness of the concept of model independence and propose a definition that recognizes that independence of models is not absolute, but comes in degrees.